Linux Audio

Check our new training course

Loading...
v6.8
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * IP Payload Compression Protocol (IPComp) - RFC3173.
  4 *
  5 * Copyright (c) 2003 James Morris <jmorris@intercode.com.au>
  6 * Copyright (c) 2003-2008 Herbert Xu <herbert@gondor.apana.org.au>
  7 *
  8 * Todo:
  9 *   - Tunable compression parameters.
 10 *   - Compression stats.
 11 *   - Adaptive compression.
 12 */
 13
 14#include <linux/crypto.h>
 15#include <linux/err.h>
 16#include <linux/list.h>
 17#include <linux/module.h>
 18#include <linux/mutex.h>
 19#include <linux/percpu.h>
 20#include <linux/slab.h>
 21#include <linux/smp.h>
 22#include <linux/vmalloc.h>
 23#include <net/ip.h>
 24#include <net/ipcomp.h>
 25#include <net/xfrm.h>
 26
 27struct ipcomp_tfms {
 28	struct list_head list;
 29	struct crypto_comp * __percpu *tfms;
 30	int users;
 31};
 32
 33static DEFINE_MUTEX(ipcomp_resource_mutex);
 34static void * __percpu *ipcomp_scratches;
 35static int ipcomp_scratch_users;
 36static LIST_HEAD(ipcomp_tfms_list);
 37
 38static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
 39{
 40	struct ipcomp_data *ipcd = x->data;
 41	const int plen = skb->len;
 42	int dlen = IPCOMP_SCRATCH_SIZE;
 43	const u8 *start = skb->data;
 44	u8 *scratch = *this_cpu_ptr(ipcomp_scratches);
 45	struct crypto_comp *tfm = *this_cpu_ptr(ipcd->tfms);
 46	int err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
 47	int len;
 48
 49	if (err)
 50		return err;
 51
 52	if (dlen < (plen + sizeof(struct ip_comp_hdr)))
 53		return -EINVAL;
 54
 55	len = dlen - plen;
 56	if (len > skb_tailroom(skb))
 57		len = skb_tailroom(skb);
 58
 59	__skb_put(skb, len);
 60
 61	len += plen;
 62	skb_copy_to_linear_data(skb, scratch, len);
 63
 64	while ((scratch += len, dlen -= len) > 0) {
 65		skb_frag_t *frag;
 66		struct page *page;
 67
 68		if (WARN_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS))
 69			return -EMSGSIZE;
 70
 71		frag = skb_shinfo(skb)->frags + skb_shinfo(skb)->nr_frags;
 72		page = alloc_page(GFP_ATOMIC);
 73
 74		if (!page)
 75			return -ENOMEM;
 76
 
 
 77		len = PAGE_SIZE;
 78		if (dlen < len)
 79			len = dlen;
 80
 81		skb_frag_fill_page_desc(frag, page, 0, len);
 
 82		memcpy(skb_frag_address(frag), scratch, len);
 83
 84		skb->truesize += len;
 85		skb->data_len += len;
 86		skb->len += len;
 87
 88		skb_shinfo(skb)->nr_frags++;
 89	}
 90
 91	return 0;
 92}
 93
 94int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
 95{
 96	int nexthdr;
 97	int err = -ENOMEM;
 98	struct ip_comp_hdr *ipch;
 99
100	if (skb_linearize_cow(skb))
101		goto out;
102
103	skb->ip_summed = CHECKSUM_NONE;
104
105	/* Remove ipcomp header and decompress original payload */
106	ipch = (void *)skb->data;
107	nexthdr = ipch->nexthdr;
108
109	skb->transport_header = skb->network_header + sizeof(*ipch);
110	__skb_pull(skb, sizeof(*ipch));
111	err = ipcomp_decompress(x, skb);
112	if (err)
113		goto out;
114
115	err = nexthdr;
116
117out:
118	return err;
119}
120EXPORT_SYMBOL_GPL(ipcomp_input);
121
122static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
123{
124	struct ipcomp_data *ipcd = x->data;
125	const int plen = skb->len;
126	int dlen = IPCOMP_SCRATCH_SIZE;
127	u8 *start = skb->data;
128	struct crypto_comp *tfm;
129	u8 *scratch;
130	int err;
131
132	local_bh_disable();
133	scratch = *this_cpu_ptr(ipcomp_scratches);
134	tfm = *this_cpu_ptr(ipcd->tfms);
135	err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
136	if (err)
137		goto out;
138
139	if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
140		err = -EMSGSIZE;
141		goto out;
142	}
143
144	memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
145	local_bh_enable();
146
147	pskb_trim(skb, dlen + sizeof(struct ip_comp_hdr));
148	return 0;
149
150out:
151	local_bh_enable();
152	return err;
153}
154
155int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
156{
157	int err;
158	struct ip_comp_hdr *ipch;
159	struct ipcomp_data *ipcd = x->data;
160
161	if (skb->len < ipcd->threshold) {
162		/* Don't bother compressing */
163		goto out_ok;
164	}
165
166	if (skb_linearize_cow(skb))
167		goto out_ok;
168
169	err = ipcomp_compress(x, skb);
170
171	if (err) {
172		goto out_ok;
173	}
174
175	/* Install ipcomp header, convert into ipcomp datagram. */
176	ipch = ip_comp_hdr(skb);
177	ipch->nexthdr = *skb_mac_header(skb);
178	ipch->flags = 0;
179	ipch->cpi = htons((u16 )ntohl(x->id.spi));
180	*skb_mac_header(skb) = IPPROTO_COMP;
181out_ok:
182	skb_push(skb, -skb_network_offset(skb));
183	return 0;
184}
185EXPORT_SYMBOL_GPL(ipcomp_output);
186
187static void ipcomp_free_scratches(void)
188{
189	int i;
190	void * __percpu *scratches;
191
192	if (--ipcomp_scratch_users)
193		return;
194
195	scratches = ipcomp_scratches;
196	if (!scratches)
197		return;
198
199	for_each_possible_cpu(i)
200		vfree(*per_cpu_ptr(scratches, i));
201
202	free_percpu(scratches);
203	ipcomp_scratches = NULL;
204}
205
206static void * __percpu *ipcomp_alloc_scratches(void)
207{
208	void * __percpu *scratches;
209	int i;
210
211	if (ipcomp_scratch_users++)
212		return ipcomp_scratches;
213
214	scratches = alloc_percpu(void *);
215	if (!scratches)
216		return NULL;
217
218	ipcomp_scratches = scratches;
219
220	for_each_possible_cpu(i) {
221		void *scratch;
222
223		scratch = vmalloc_node(IPCOMP_SCRATCH_SIZE, cpu_to_node(i));
224		if (!scratch)
225			return NULL;
226		*per_cpu_ptr(scratches, i) = scratch;
227	}
228
229	return scratches;
230}
231
232static void ipcomp_free_tfms(struct crypto_comp * __percpu *tfms)
233{
234	struct ipcomp_tfms *pos;
235	int cpu;
236
237	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
238		if (pos->tfms == tfms)
239			break;
240	}
241
242	WARN_ON(list_entry_is_head(pos, &ipcomp_tfms_list, list));
243
244	if (--pos->users)
245		return;
246
247	list_del(&pos->list);
248	kfree(pos);
249
250	if (!tfms)
251		return;
252
253	for_each_possible_cpu(cpu) {
254		struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
255		crypto_free_comp(tfm);
256	}
257	free_percpu(tfms);
258}
259
260static struct crypto_comp * __percpu *ipcomp_alloc_tfms(const char *alg_name)
261{
262	struct ipcomp_tfms *pos;
263	struct crypto_comp * __percpu *tfms;
264	int cpu;
265
266
267	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
268		struct crypto_comp *tfm;
269
270		/* This can be any valid CPU ID so we don't need locking. */
271		tfm = this_cpu_read(*pos->tfms);
272
273		if (!strcmp(crypto_comp_name(tfm), alg_name)) {
274			pos->users++;
275			return pos->tfms;
276		}
277	}
278
279	pos = kmalloc(sizeof(*pos), GFP_KERNEL);
280	if (!pos)
281		return NULL;
282
283	pos->users = 1;
284	INIT_LIST_HEAD(&pos->list);
285	list_add(&pos->list, &ipcomp_tfms_list);
286
287	pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
288	if (!tfms)
289		goto error;
290
291	for_each_possible_cpu(cpu) {
292		struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
293							    CRYPTO_ALG_ASYNC);
294		if (IS_ERR(tfm))
295			goto error;
296		*per_cpu_ptr(tfms, cpu) = tfm;
297	}
298
299	return tfms;
300
301error:
302	ipcomp_free_tfms(tfms);
303	return NULL;
304}
305
306static void ipcomp_free_data(struct ipcomp_data *ipcd)
307{
308	if (ipcd->tfms)
309		ipcomp_free_tfms(ipcd->tfms);
310	ipcomp_free_scratches();
311}
312
313void ipcomp_destroy(struct xfrm_state *x)
314{
315	struct ipcomp_data *ipcd = x->data;
316	if (!ipcd)
317		return;
318	xfrm_state_delete_tunnel(x);
319	mutex_lock(&ipcomp_resource_mutex);
320	ipcomp_free_data(ipcd);
321	mutex_unlock(&ipcomp_resource_mutex);
322	kfree(ipcd);
323}
324EXPORT_SYMBOL_GPL(ipcomp_destroy);
325
326int ipcomp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
327{
328	int err;
329	struct ipcomp_data *ipcd;
330	struct xfrm_algo_desc *calg_desc;
331
332	err = -EINVAL;
333	if (!x->calg) {
334		NL_SET_ERR_MSG(extack, "Missing required compression algorithm");
335		goto out;
336	}
337
338	if (x->encap) {
339		NL_SET_ERR_MSG(extack, "IPComp is not compatible with encapsulation");
340		goto out;
341	}
342
343	err = -ENOMEM;
344	ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
345	if (!ipcd)
346		goto out;
347
348	mutex_lock(&ipcomp_resource_mutex);
349	if (!ipcomp_alloc_scratches())
350		goto error;
351
352	ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
353	if (!ipcd->tfms)
354		goto error;
355	mutex_unlock(&ipcomp_resource_mutex);
356
357	calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
358	BUG_ON(!calg_desc);
359	ipcd->threshold = calg_desc->uinfo.comp.threshold;
360	x->data = ipcd;
361	err = 0;
362out:
363	return err;
364
365error:
366	ipcomp_free_data(ipcd);
367	mutex_unlock(&ipcomp_resource_mutex);
368	kfree(ipcd);
369	goto out;
370}
371EXPORT_SYMBOL_GPL(ipcomp_init_state);
372
373MODULE_LICENSE("GPL");
374MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
375MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
v6.2
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * IP Payload Compression Protocol (IPComp) - RFC3173.
  4 *
  5 * Copyright (c) 2003 James Morris <jmorris@intercode.com.au>
  6 * Copyright (c) 2003-2008 Herbert Xu <herbert@gondor.apana.org.au>
  7 *
  8 * Todo:
  9 *   - Tunable compression parameters.
 10 *   - Compression stats.
 11 *   - Adaptive compression.
 12 */
 13
 14#include <linux/crypto.h>
 15#include <linux/err.h>
 16#include <linux/list.h>
 17#include <linux/module.h>
 18#include <linux/mutex.h>
 19#include <linux/percpu.h>
 20#include <linux/slab.h>
 21#include <linux/smp.h>
 22#include <linux/vmalloc.h>
 23#include <net/ip.h>
 24#include <net/ipcomp.h>
 25#include <net/xfrm.h>
 26
 27struct ipcomp_tfms {
 28	struct list_head list;
 29	struct crypto_comp * __percpu *tfms;
 30	int users;
 31};
 32
 33static DEFINE_MUTEX(ipcomp_resource_mutex);
 34static void * __percpu *ipcomp_scratches;
 35static int ipcomp_scratch_users;
 36static LIST_HEAD(ipcomp_tfms_list);
 37
 38static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
 39{
 40	struct ipcomp_data *ipcd = x->data;
 41	const int plen = skb->len;
 42	int dlen = IPCOMP_SCRATCH_SIZE;
 43	const u8 *start = skb->data;
 44	u8 *scratch = *this_cpu_ptr(ipcomp_scratches);
 45	struct crypto_comp *tfm = *this_cpu_ptr(ipcd->tfms);
 46	int err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
 47	int len;
 48
 49	if (err)
 50		return err;
 51
 52	if (dlen < (plen + sizeof(struct ip_comp_hdr)))
 53		return -EINVAL;
 54
 55	len = dlen - plen;
 56	if (len > skb_tailroom(skb))
 57		len = skb_tailroom(skb);
 58
 59	__skb_put(skb, len);
 60
 61	len += plen;
 62	skb_copy_to_linear_data(skb, scratch, len);
 63
 64	while ((scratch += len, dlen -= len) > 0) {
 65		skb_frag_t *frag;
 66		struct page *page;
 67
 68		if (WARN_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS))
 69			return -EMSGSIZE;
 70
 71		frag = skb_shinfo(skb)->frags + skb_shinfo(skb)->nr_frags;
 72		page = alloc_page(GFP_ATOMIC);
 73
 74		if (!page)
 75			return -ENOMEM;
 76
 77		__skb_frag_set_page(frag, page);
 78
 79		len = PAGE_SIZE;
 80		if (dlen < len)
 81			len = dlen;
 82
 83		skb_frag_off_set(frag, 0);
 84		skb_frag_size_set(frag, len);
 85		memcpy(skb_frag_address(frag), scratch, len);
 86
 87		skb->truesize += len;
 88		skb->data_len += len;
 89		skb->len += len;
 90
 91		skb_shinfo(skb)->nr_frags++;
 92	}
 93
 94	return 0;
 95}
 96
 97int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
 98{
 99	int nexthdr;
100	int err = -ENOMEM;
101	struct ip_comp_hdr *ipch;
102
103	if (skb_linearize_cow(skb))
104		goto out;
105
106	skb->ip_summed = CHECKSUM_NONE;
107
108	/* Remove ipcomp header and decompress original payload */
109	ipch = (void *)skb->data;
110	nexthdr = ipch->nexthdr;
111
112	skb->transport_header = skb->network_header + sizeof(*ipch);
113	__skb_pull(skb, sizeof(*ipch));
114	err = ipcomp_decompress(x, skb);
115	if (err)
116		goto out;
117
118	err = nexthdr;
119
120out:
121	return err;
122}
123EXPORT_SYMBOL_GPL(ipcomp_input);
124
125static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
126{
127	struct ipcomp_data *ipcd = x->data;
128	const int plen = skb->len;
129	int dlen = IPCOMP_SCRATCH_SIZE;
130	u8 *start = skb->data;
131	struct crypto_comp *tfm;
132	u8 *scratch;
133	int err;
134
135	local_bh_disable();
136	scratch = *this_cpu_ptr(ipcomp_scratches);
137	tfm = *this_cpu_ptr(ipcd->tfms);
138	err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
139	if (err)
140		goto out;
141
142	if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
143		err = -EMSGSIZE;
144		goto out;
145	}
146
147	memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
148	local_bh_enable();
149
150	pskb_trim(skb, dlen + sizeof(struct ip_comp_hdr));
151	return 0;
152
153out:
154	local_bh_enable();
155	return err;
156}
157
158int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
159{
160	int err;
161	struct ip_comp_hdr *ipch;
162	struct ipcomp_data *ipcd = x->data;
163
164	if (skb->len < ipcd->threshold) {
165		/* Don't bother compressing */
166		goto out_ok;
167	}
168
169	if (skb_linearize_cow(skb))
170		goto out_ok;
171
172	err = ipcomp_compress(x, skb);
173
174	if (err) {
175		goto out_ok;
176	}
177
178	/* Install ipcomp header, convert into ipcomp datagram. */
179	ipch = ip_comp_hdr(skb);
180	ipch->nexthdr = *skb_mac_header(skb);
181	ipch->flags = 0;
182	ipch->cpi = htons((u16 )ntohl(x->id.spi));
183	*skb_mac_header(skb) = IPPROTO_COMP;
184out_ok:
185	skb_push(skb, -skb_network_offset(skb));
186	return 0;
187}
188EXPORT_SYMBOL_GPL(ipcomp_output);
189
190static void ipcomp_free_scratches(void)
191{
192	int i;
193	void * __percpu *scratches;
194
195	if (--ipcomp_scratch_users)
196		return;
197
198	scratches = ipcomp_scratches;
199	if (!scratches)
200		return;
201
202	for_each_possible_cpu(i)
203		vfree(*per_cpu_ptr(scratches, i));
204
205	free_percpu(scratches);
206	ipcomp_scratches = NULL;
207}
208
209static void * __percpu *ipcomp_alloc_scratches(void)
210{
211	void * __percpu *scratches;
212	int i;
213
214	if (ipcomp_scratch_users++)
215		return ipcomp_scratches;
216
217	scratches = alloc_percpu(void *);
218	if (!scratches)
219		return NULL;
220
221	ipcomp_scratches = scratches;
222
223	for_each_possible_cpu(i) {
224		void *scratch;
225
226		scratch = vmalloc_node(IPCOMP_SCRATCH_SIZE, cpu_to_node(i));
227		if (!scratch)
228			return NULL;
229		*per_cpu_ptr(scratches, i) = scratch;
230	}
231
232	return scratches;
233}
234
235static void ipcomp_free_tfms(struct crypto_comp * __percpu *tfms)
236{
237	struct ipcomp_tfms *pos;
238	int cpu;
239
240	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
241		if (pos->tfms == tfms)
242			break;
243	}
244
245	WARN_ON(list_entry_is_head(pos, &ipcomp_tfms_list, list));
246
247	if (--pos->users)
248		return;
249
250	list_del(&pos->list);
251	kfree(pos);
252
253	if (!tfms)
254		return;
255
256	for_each_possible_cpu(cpu) {
257		struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
258		crypto_free_comp(tfm);
259	}
260	free_percpu(tfms);
261}
262
263static struct crypto_comp * __percpu *ipcomp_alloc_tfms(const char *alg_name)
264{
265	struct ipcomp_tfms *pos;
266	struct crypto_comp * __percpu *tfms;
267	int cpu;
268
269
270	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
271		struct crypto_comp *tfm;
272
273		/* This can be any valid CPU ID so we don't need locking. */
274		tfm = this_cpu_read(*pos->tfms);
275
276		if (!strcmp(crypto_comp_name(tfm), alg_name)) {
277			pos->users++;
278			return pos->tfms;
279		}
280	}
281
282	pos = kmalloc(sizeof(*pos), GFP_KERNEL);
283	if (!pos)
284		return NULL;
285
286	pos->users = 1;
287	INIT_LIST_HEAD(&pos->list);
288	list_add(&pos->list, &ipcomp_tfms_list);
289
290	pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
291	if (!tfms)
292		goto error;
293
294	for_each_possible_cpu(cpu) {
295		struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
296							    CRYPTO_ALG_ASYNC);
297		if (IS_ERR(tfm))
298			goto error;
299		*per_cpu_ptr(tfms, cpu) = tfm;
300	}
301
302	return tfms;
303
304error:
305	ipcomp_free_tfms(tfms);
306	return NULL;
307}
308
309static void ipcomp_free_data(struct ipcomp_data *ipcd)
310{
311	if (ipcd->tfms)
312		ipcomp_free_tfms(ipcd->tfms);
313	ipcomp_free_scratches();
314}
315
316void ipcomp_destroy(struct xfrm_state *x)
317{
318	struct ipcomp_data *ipcd = x->data;
319	if (!ipcd)
320		return;
321	xfrm_state_delete_tunnel(x);
322	mutex_lock(&ipcomp_resource_mutex);
323	ipcomp_free_data(ipcd);
324	mutex_unlock(&ipcomp_resource_mutex);
325	kfree(ipcd);
326}
327EXPORT_SYMBOL_GPL(ipcomp_destroy);
328
329int ipcomp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
330{
331	int err;
332	struct ipcomp_data *ipcd;
333	struct xfrm_algo_desc *calg_desc;
334
335	err = -EINVAL;
336	if (!x->calg) {
337		NL_SET_ERR_MSG(extack, "Missing required compression algorithm");
338		goto out;
339	}
340
341	if (x->encap) {
342		NL_SET_ERR_MSG(extack, "IPComp is not compatible with encapsulation");
343		goto out;
344	}
345
346	err = -ENOMEM;
347	ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
348	if (!ipcd)
349		goto out;
350
351	mutex_lock(&ipcomp_resource_mutex);
352	if (!ipcomp_alloc_scratches())
353		goto error;
354
355	ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
356	if (!ipcd->tfms)
357		goto error;
358	mutex_unlock(&ipcomp_resource_mutex);
359
360	calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
361	BUG_ON(!calg_desc);
362	ipcd->threshold = calg_desc->uinfo.comp.threshold;
363	x->data = ipcd;
364	err = 0;
365out:
366	return err;
367
368error:
369	ipcomp_free_data(ipcd);
370	mutex_unlock(&ipcomp_resource_mutex);
371	kfree(ipcd);
372	goto out;
373}
374EXPORT_SYMBOL_GPL(ipcomp_init_state);
375
376MODULE_LICENSE("GPL");
377MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
378MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");